Raising the Speed Limit: U.S. Economic Growth in the ......1990–91 recession (Bureau of Labor Statistics, 2000). ... plays as a source of economic growth. 3 For the period 1959–73,

Raising the Speed Limit: U.S. Economic Growth in the Information Age

THE CONTINUED STRENGTH and vitality of the U.S. economy continue toastonish economic forecasters.1 A consensus is now emerging thatsomething fundamental has changed, with “new economy” proponentspointing to information technology (IT) as the causal factor behind the strong performance. In this view, technology is profoundly alter-ing the nature of business, leading to permanently higher productivitygrowth throughout the economy. Skeptics remain, however, arguing that the recent success reflects a series of favorable, but temporary,shocks. This argument is buttressed by the view that the U.S. econ-

125

D A L E W. J O R G E N S O NHarvard University

K E V I N J . S T I R O HFederal Reserve Bank of New York

We are indebted to Mun Ho for his comments and assistance with the industry and labor data. We are also grateful to Robert Arnold of the Congressional Budget Office forhelpful comments and discussions of that agency’s results and methods, and to BruceGrimm and David Wasshausen of the Bureau of Economic Analysis for details on their agency’s investment data and prices. Our thanks are due also to Erwin Diewert, RobertGordon, Stephen Oliner, Daniel Sichel, and Kun-Young Yun, as well as to seminar partici-pants at the Brookings Panel, the Federal Reserve Bank of New York, and the FederalReserve Board for helpful comments and advice. David Fiore provided excellent researchassistance. The views expressed in this paper are those of the authors only and do not nec-essarily reflect the views of the Federal Reserve Bank of New York or the Federal ReserveSystem.

1. Labor productivity growth for the business sector averaged 2.7 percent per year dur-ing 1995–99. These four years recorded the four fastest annual productivity growth rates inthe 1990s, except for a temporary jump of 4.3 percent in 1992 as the economy exited the1990–91 recession (Bureau of Labor Statistics, 2000).

9573—04 BPEA Jorgenson/Stiroh 7/21/00 10:22 Page 125

omy behaves rather differently than envisioned by the “new economy”advocates.2

Productivity growth, capital accumulation, and the impact of technol-ogy were topics once reserved for academic debates, but the recent successof the U.S. economy has moved them into popular discussion. This paperemploys well-tested and familiar methods to analyze important new infor-mation made available by the recent benchmark revision of the U.S.national income and product accounts (NIPAs). We document the casefor raising the speed limit: for an upward revision of intermediate-termprojections of future growth to reflect the latest data and trends.

The late 1990s were exceptional in comparison with the growth expe-rience of the U.S. economy over the past quarter century as a whole.Although growth rates have not yet returned to those of the golden age ofthe U.S. economy in the 1960s, the data nonetheless clearly reveal aremarkable transformation. Rapid declines in the prices of computers andsemiconductors are well known and carefully documented, and evidence isaccumulating that similar declines are taking place in the prices of soft-ware and communications equipment. Unfortunately, the empirical recordis seriously incomplete, and therefore much remains to be done beforedefinitive quantitative assessments can be made about the complete role ofthese high-technology assets.

Despite the limitations of the available data, the mechanisms under-lying the structural transformation of the U.S. economy are readily appar-ent. As an illustration, consider the increasing role that computer hardwareplays as a source of economic growth.3 For the period 1959–73, computerinputs contributed less than 0.1 percentage point to annual U.S. economicgrowth. Since 1973, however, the price of computers has fallen at a his-torically unprecedented rate, and firms and households, following a basicprinciple of economics, have substituted toward these relatively cheaper

126 Brookings Papers on Economic Activity, 1:2000

2. Stiroh (1999) critiques alternative views on the new economy, Triplett (1999)examines some data issues in the new economy debate, and Gordon (1999b) provides anoften-cited rebuttal of the new economy thesis.

3. Our work on computers builds on the path-breaking research of Oliner and Sichel(1994, 2000) and Sichel (1997, 1999) and our own earlier results, reported in Jorgensonand Stiroh (1995, 1999, 2000) and Stiroh (1998a). Other valuable work on computersincludes that of Haimowitz (1998), Kiley (1999), and Whelan (2000). Gordon (1999a) pro-vides a historical perspective on the sources of U.S. economic growth, and Brynjolfsson andYang (1996) review the microeconomic evidence on computers and productivity.


inputs. Since 1995 the price decline for computers has accelerated, reach-ing nearly 28 percent per year from 1995 to 1998. In response, invest-ment in computers has exploded, and the growth contribution of computerhardware has increased more than fivefold, to 0.46 percentage point peryear in the late 1990s.4 Software and communications equipment, twoother types of IT assets, contributed an additional 0.30 percentage pointper year for 1995–98. Preliminary estimates through 1999 reveal furtherincreases in these contributions for all three high-technology assets.

Next, consider the acceleration of average labor productivity (ALP)growth in the 1990s. After a twenty-year slowdown dating from the early1970s, ALP grew 2.4 percent per year during 1995–98, more than a per-centage point faster than during 1990–95.5 A detailed decompositionshows that capital deepening, the direct consequence of price-induced sub-stitution and rapid investment, added 0.49 percentage point to ALPgrowth. Faster total factor productivity (TFP) growth contributed an addi-tional 0.63 percentage point, partly reflecting technical change in the pro-duction of computers and the resulting acceleration in their price decline.Meanwhile, slowing growth in labor quality retarded ALP growth by 0.12 percentage point relative to the early 1990s, as employers exhaustedthe pool of available workers.

TFP growth had been an anemic 0.34 percent per year for 1973–95 butaccelerated to 0.99 percent for 1995–98. After more than twenty years ofsluggish TFP growth, four of the five years ending in 1998 saw growthrates near 1 percent. It could be argued that this represents a new paradigm.In this view, the diffusion of IT improves business practices, generatesspillover benefits, and raises productivity throughout the economy. If thistrend is sustainable, it could revive the optimistic expectations of the 1960sand overcome the pessimism of the “Age of Diminished Expectations,”as an influential book at the beginning of the 1990s called the era of theproductivity slowdown.6

Dale W. Jorgenson and Kevin J. Stiroh 127

4. See Baily and Gordon (1988), Stiroh (1998a), Jorgenson and Stiroh (1999), and U.S.Department of Commerce (1999) for earlier discussions of relative price changes and inputsubstitution in the high-technology areas.

5. The Bureau of Labor Statistics’ (2000) estimates for the business sector show a simi-lar increase, from 1.6 percent per year for 1990–95 to 2.6 percent for 1995–98. See Coun-cil of Economic Advisers (2000, p. 35) for a comparison of productivity growth at variouspoints in the economic expansions of the 1960s, 1980s, and 1990s.

6. Krugman (1990).


A closer look at the data, however, shows that gains in TFP growth canbe traced in large part to IT industries, which produce computers, semi-conductors, and other high-technology gear. The evidence is equally clearthat computer-using industries such as finance, insurance, and real estate(FIRE) and other services have continued to lag in productivity growth.Reconciling the massive investment in high technology with the relativelyslow productivity growth observed in service industries remains an impor-tant task for proponents of the “new economy” position.7

What does this imply for the future? The sustainability of growth inlabor productivity is the key issue for future growth projections. For somepurposes, the distinctions among capital accumulation and growth in laborquality and TFP may not matter, so long as ALP growth can be expected tocontinue. It is sustainable labor productivity gains, after all, that ultimatelydrive long-run growth and raise living standards.

In this respect, the recent experience provides grounds for caution,since much depends on productivity gains in high-technology industries.Ongoing technological advances in these industries have been a directsource of improvement in TFP growth, as well as an indirect source ofmore-rapid capital deepening. The sustainability of this growth, there-fore, hinges critically on the future pace of technological progress in theseindustries. As measured by relative price changes, this progress has accel-erated recently: the 28 percent per year decline in computer prices during1995–98, mentioned above, compares with only a 15 percent per yeardecline in 1990–95. There is no guarantee, of course, of continued pro-ductivity gains and price declines of this magnitude. Nonetheless, as longas high-technology industries maintain the ability to innovate and improvetheir productivity at rates comparable even to their long-term averages, rel-ative prices will fall, and the virtuous circle of an investment-led expansionwill continue.8

Finally, we argue that the rewards from new technology accrue to thedirect participants: first, to the innovating industries producing the high-


7. See Gullickson and Harper (1999), Jorgenson and Stiroh (2000), and below forindustry-level analyses.

8. There is no consensus, however, that technical progress in computer and semicon-ductor production is slowing. According to Lawrence M. Fisher (“New Era Approaches:Gigabyte Chips.” New York Times, February 7, 2000, p. C8), chip processing speed contin-ues to increase rapidly. Moreover, the product cycle is accelerating as new processors arebeing brought to market more quickly.


technology assets, and second, to the industries that restructure to imple-ment the latest technology. There is little evidence of spillovers fromproduction of IT to the industries that use this technology. Indeed, many ofthe industries that use IT most intensively, such as FIRE and otherservices, show high rates of substitution of IT for other inputs but rela-tively low rates of productivity growth. In part, this may reflect problemsin measuring the output of these industries, but the empirical record pro-vides little support for the “new economy” picture of spillovers cascadingfrom IT producers onto users of this technology.9

This paper is organized as follows. The next section describes ourmethodology for quantifying the sources of U.S. economic growth. Wepresent results for the period 1959–98 and focus on the period of the late1990s. We then explore the implications of the recent experience for futuregrowth, comparing our results with recent estimates by the CongressionalBudget Office, the Council of Economic Advisers, and the Office of Man-agement and Budget. Finally, we move beyond the aggregate data to quan-tify productivity growth at the industry level. Using methodology firstintroduced by Evsey Domar, we consider the impact of IT on aggregateproductivity.

The Recent U.S. Growth Experience

The U.S. economy has undergone a remarkable transformation in recentyears, with growth in output, labor productivity, and TFP all acceleratingsince the mid-1990s. This growth resurgence has led to a widening debateabout the sources of this growth and whether profound changes are tak-ing place in the structure of the economy. Proponents of the view that weare in a “new economy” trace the increased growth to developments inIT, especially the rapid commercialization of the Internet, which theyclaim are fundamentally changing economic activity. “Old economy”advocates focus on the lackluster performance during the first half of the1990s, the increase in labor force participation and the rapid decline inunemployment since 1993, and the recent investment boom.


9. See Dean (1999) and Gullickson and Harper (1999) for the Bureau of Labor Statistics’perspective on measurement error; Triplett and Bosworth (2000) provide an overview ofmeasuring output in the service industries.


Our objective here is to quantify the sources of the recent surge in U.S.economic growth, using new information made available by the bench-mark revision of the NIPAs released in October 1999 by the Bureau ofEconomic Analysis (BEA).10 We then consider the implications of ourresults for intermediate-term projections of U.S. economic growth. Wegive special attention to the rapid escalation in growth rates in the officialprojections, such as those by the Congressional Budget Office (CBO) andthe Council of Economic Advisers (CEA). The CBO projections are par-ticularly suitable for our purposes, since they are widely disseminated,are well documented, and represent best practice. We do not focus on theissue of inflation and do not comment on potential implications for mon-etary policy.

Sources of Economic Growth

Our methodology is based on the production possibilities frontier intro-duced by Dale Jorgenson and first employed by Jorgenson and ZviGriliches.11 This captures substitutions among outputs of investment andconsumption goods, as well as between inputs of capital and labor. Weidentify IT with investments in computers, software, and communica-tions equipment, as well as consumption of computers and software as out-puts. The service flows from these assets are also inputs. The aggregateproduction function employed by Robert Solow and, more recently, byJeremy Greenwood, Zvi Hercowitz, and Per Krusell, is an alternative toour model.12 In their approach a single output is expressed as a functionof capital and labor inputs. This implicitly assumes, however, that invest-ments in IT are perfect substitutes for other outputs, so that relative pricesdo not change.

Our methodology is essential in order to capture two important factsabout which there is general agreement. The first is that prices of com-puters have declined drastically relative to the prices of other investmentgoods. The second is that this rate of decline has recently accelerated. Inaddition, estimates of investment in software, now available in the NIPAs,are comparable to investment in hardware. The new data show that theprice of software has fallen relative to the prices of other investment goods,


10. Data are available at www.bea.doc.gov.11. Jorgenson (1966); Jorgenson and Griliches (1967).12. Solow (1957, 1960); Greenwood, Hercowitz, and Krusell (1997).


but more slowly than the price of hardware. We examine the estimates ofsoftware investment in some detail in order to assess the role of softwarein recent economic growth. Finally, we consider investment in communi-cations equipment, which shares many of the technological features ofcomputer hardware.

THE PRODUCTION POSSIBILITIES FRONTIER. Aggregate output Yt consistsof investment goods It and consumption goods Ct. These outputs are pro-duced from aggregate input Xt, consisting of capital services Kt and laborservices Lt. We represent productivity as a Hicks-neutral augmentation Atof aggregate input:13

(1)

The outputs of investment and consumption goods and the inputs of capital and labor services are themselves aggregates, each with manysubcomponents.

Under the assumptions of competitive product and factor markets andconstant returns to scale, growth accounting gives the share-weightedgrowth of outputs as the sum of the share-weighted growth of inputs andgrowth in TFP:

(2)

where w� I,t is investment’s average share of nominal output, w�C,t isconsumption’s average share of nominal output, v�K,t is capital’s averageshare of nominal income, v�L,t is labor’s average share of nominal income,w� I,t + w�C,t = v�K,t + v�L,t = 1, and � refers to a first difference. Note that wereserve the term TFP for the augmentation factor in equation 1.

Equation 2 enables us to identify the contributions of outputs as wellas of inputs to economic growth. For example, we can quantify the con-tributions of different investments, such as computers, software, and com-munications equipment, to the growth of output by decomposing thegrowth of investment among its subcomponents. Similarly, we can quan-tify the contributions of different types of consumption, such as servicesfrom computers and software, by decomposing the growth of consump-

,, , , ,w I w C v K v L AI t t C t t K t t L t t t∆ ∆ ∆ ∆ ∆ln ln ln ln ln+ = + +

, , .Y I C A X K Lt t t t t( ) = ⋅ ( )


13. It would be a straightforward change to make technology labor-augmenting, orHarrod-neutral, so that the production possibilities frontier could be written as Y(I, C) =X(K, AL). Also, there is no need to assume that inputs and outputs are separable, but thissimplifies our notation.


tion. As shown by Jorgenson and Kevin Stiroh,14 both computer investmentand consumption of IT made important contributions to U.S. economicgrowth in the 1990s. We also consider the output contributions of softwareand communications equipment as distinct high-technology assets. Simi-larly, we decompose the contribution of capital input to isolate the impactof computers, software, and communications equipment on input growth.

Rearranging equation 2 enables us to present results in terms of growthin ALP, defined as yt = Yt /Ht , where Yt is output, defined as an aggregate ofconsumption and investment goods, and kt = Kt /Ht is the ratio of capitalservices to hours worked:

(3)

This gives the familiar allocation of ALP growth among three factors.The first is capital deepening, the growth in capital services per hourworked. Capital deepening makes workers more productive by providingmore capital for each hour of work, and it raises the growth of ALP inproportion to the share of capital. The second term is the improvement inlabor quality, defined as the difference between growth rates of labor inputand hours worked. Reflecting the rising proportion of hours supplied byworkers with higher marginal products, an improvement in labor qualityraises ALP growth in proportion to labor’s share. The third factor is TFPgrowth, which increases ALP growth on a point-for-point basis.

COMPUTERS, SOFTWARE, AND COMMUNICATIONS EQUIPMENT. We nowconsider the impact of investment in computers, software, and communi-cations equipment on economic growth. For this purpose we must care-fully distinguish the use of IT from the production of IT.15 For example,computers themselves are an output of one industry (commercial andindustrial machinery), and computing services are inputs into other indus-tries (computer-using industries such as trade, FIRE, and other services).

Massive increases in computing power, like those currently experiencedby the U.S. economy, therefore have two effects on growth. First, as theproduction of computers improves and becomes more efficient, morecomputing power is being produced from the same inputs. This raisesoverall productivity in the computer-producing industry and contributes to

., ,∆ ∆ ∆ ∆ ∆ln ln ln ln lny v k v L H At K t t L t t t t= + −( ) +


14. Jorgenson and Stiroh (1999); Gordon (1999b).15. Baily and Gordon (1988), Griliches (1992), Stiroh (1998a), Jorgenson and Stiroh

(1999), Whelan (2000), and Oliner and Sichel (2000) discuss the impact of investment incomputers from these two perspectives.


TFP growth for the economy as a whole. Labor productivity also growsat both the industry and the aggregate levels.16

Second, the rapid accumulation of computers leads to growth of com-puting power as an input in computer-using industries. Since labor isworking with more and better computer equipment, this investmentincreases labor productivity. If the contributions to output are captured bythe effect of capital deepening, TFP growth is unaffected. As Martin Bailyand Robert Gordon remark, “there is no shift in the user firm’s produc-tion function,”17 and thus no gain in TFP. Increasing deployment of com-puters increases TFP only if there are spillovers from the production ofcomputers to production in the computer-using industries, or if there aremeasurement problems associated with the new inputs.

We conclude that rapid growth in computing power affects aggregateoutput through both TFP growth and capital deepening. Progress in thetechnology of computer production contributes to growth in TFP and ALPat the aggregate level. The accumulation of computing power in computer-using industries reflects the substitution of computers for other inputs andleads to growth in ALP. In the absence of spillovers, this growth does notcontribute to growth in TFP.

The remainder of this section provides empirical estimates of the vari-ables in equations 1 through 3. We then employ equations 2 and 3 to quan-tify the sources of growth of output and ALP for 1959–98 and varioussubperiods.

Output

Our output data are based on the most recent benchmark revision ofthe NIPAs.18 Real output Yt is measured in chained 1996 dollars, and PY,tis the corresponding implicit deflator. Our output concept is similar, butnot identical, to one used in the Bureau of Labor Statistics (BLS) produc-tivity program. Like the BLS, we exclude the government sector, but


16. Triplett (1996) points out that much of the decline in computer prices reflects fallingsemiconductor prices. If all inputs are correctly measured for quality change, therefore,much of the TFP gain in computer production is rightly pushed back to TFP gains in semi-conductor production, since semiconductors are a major intermediate input in the produc-tion of computers. See Flamm (1993) for early estimates on semiconductor prices. Weaddress this issue further below.

17. Baily and Gordon (1988, p. 378).18. See appendix A for details on our source data and methodology for output estimates.


unlike the BLS we include imputations for the service flow from consumerdurables and owner-occupied housing. These imputations are necessaryto preserve comparability between durables and housing, and they enableus to capture the important impact of IT on households.

Our estimate of current dollar, private output in 1998 is $8.013 trillion,including imputations of $740 billion that primarily reflect services of con-sumer durables.19 Real output growth was 3.63 percent per year for the fullperiod 1959–98, compared with 3.36 percent for the official GDP series.This difference reflects both our imputations and our exclusion of the gov-ernment sectors in the NIPA data. Appendix table A1 presents the currentdollar value and corresponding price index of total private output and ofeach major category of IT assets: investment in computers Ic, investment insoftware Is, investment in communications equipment Im, consumption ofcomputers and software Cc, and the imputed service flow from consumers’computers and software Dc.

The most striking feature of these data is the enormous price decline forcomputer investment: 18 percent per year from 1960 to 1995 (figure 1).Since 1995 this decline has accelerated to 27.6 percent per year, as notedabove. By contrast, the relative price of software investment has been flatfor much of the period and only began to fall in the late 1980s. The priceof communications equipment behaves in a manner similar to the softwareprice, whereas the price of consumption of computers and software showsdeclines similar to that of computer investment. The top panel of table 1summarizes the growth rates of prices and quantities for major outputcategories for 1990–95 and for 1995–98.

In terms of current dollar output, investment in software was the largestIT asset in 1998, followed by investment in computers, with investmentin communications equipment a close third (figure 2). Although businessinvestments in computers, software, and communications equipment areby far the largest categories, households have spent more than $20 billionper year on computers and software since 1995, generating a service flowof comparable magnitude.


19. Current dollar GDP in the NIPAs in 1998 was $8.759.9 trillion. Our estimate differsbecause of total imputations ($740 billion), our exclusion of the general government andgovernment enterprise sectors ($972 billion and $128 billion, respectively), and our exclu-sion of certain retail taxes ($376 billion).


Capital Stock and Capital Services

Here we describe our capital estimates for the U.S. economy from 1959to 1998.20 We begin with investment data from the BEA, estimate capitalstocks using the perpetual inventory method, and aggregate capital stocksusing rental prices as weights. This approach, originated by Jorgenson andGriliches,21 is based on the identification of rental prices with marginalproducts of different types of capital. Our estimates of these prices incor-porate differences in asset prices, service lives and depreciation rates, andthe tax treatment of capital incomes.22


20. See appendix B for details on the theory, source data, and methodology for our cap-ital estimates.

21. Jorgenson and Griliches (1967).22. Jorgenson (1996) provides a recent discussion of our model of capital as a factor of

production. BLS (1983) describes the version of this model employed in the official pro-ductivity statistics. Hulten (2000) provides a review of the specific features of this method-ology for measuring capital input and the link to economic theory.

1,000

100

10

1

1965 1970 1975 1980 1985 1990 1995

Computer investment

Software investment

Communications investment

Computer and software consumer durable services

Computer and software consumption

Index (1996 = 1.0; log scale)

Figure 1. Relative Prices of Information Technology Outputs, 1960–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. All price indexes are relative to the output price index.


We refer to the difference between growth in capital services andgrowth in the capital stock as the growth in capital quality qK,t; this repre-sents substitution toward assets with higher marginal products.23 Forexample, the shift toward IT increases the quality of capital, since com-puters, software, and communications equipment are assets with relativelyhigh marginal products. Capital stock estimates, like those originallyemployed by Solow,24 fail to account for this increase in quality.

We employ a broad definition of capital, including tangible assets suchas equipment and structures, as well as consumer durables, land, and


23. More precisely, growth in capital quality is defined as the difference between thegrowth in capital services and the growth in the average of the current and lagged capitalstock. Appendix B provides details. We use a geometric depreciation rate for all repro-ducible assets, so that our estimates are not identical to the wealth estimates published inBEA (1998b).

24. Solow (1957).

Table 1. Average Growth Rates of Prices and Quantities for Selected Outputs andInputs, 1990–98 Percent

1990–95 1995–98

Type of output or input Prices Quantities Prices Quantities

OutputsPrivate domestic output (Y) 1.70 2.74 1.37 4.73

Other (Yn) 2.01 2.25 2.02 3.82Computer and software consumption (Cc) –21.50 38.67 –36.93 49.26Computer investment (Ic) –14.59 24.89 –27.58 38.08Software investment (Is) –1.41 11.59 –2.16 15.18Communications investment (Im) –1.50 6.17 –1.73 12.79Computer and software consumer durable –19.34 34.79 –28.62 44.57

services (Dc)

InputsTotal capital services (K) 0.60 2.83 2.54 4.80

Other (Kn) 1.00 1.78 4.20 2.91Computer capital (Kc) –10.59 18.16 –20.09 34.10Software capital (Ks) –2.07 13.22 –0.87 13.00Communications capital (Km) 3.10 4.31 –7.09 7.80

Total consumption services (D) 1.98 2.91 –0.67 5.39Other (Dn) 2.55 2.07 0.54 3.73Computer and software consumer durable –19.34 34.79 –28.62 44.57

services (Dc)Labor (L) 2.92 2.01 2.80 2.81

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.


inventories. We estimate a service flow from the installed stock of con-sumer durables, which enters our measures of both output and input. It isessential to include this service flow, since a steadily rising proportion isassociated with purchases of IT by the household sector. In order to cap-ture the impact of IT on U.S. economic growth, investments by the busi-ness and household sectors as well as the services of the resulting capitalstocks must be included.

Our estimate of the capital stock is $26 trillion in 1997, substantiallylarger than the $17.3 trillion in fixed private capital estimated by theBEA.25 This difference reflects our inclusion of consumer durables, inven-tories, and land. Our estimates of the capital stock for comparable cate-gories of assets are quite similar to those of the BEA. Our estimate of fixedprivate capital in 1997, for example, is $16.8 trillion, almost the same asthe BEA’s. Similarly, our estimate of the stock of consumer durables is$2.9 trillion, whereas the BEA’s estimate is $2.5 trillion. The remainingdiscrepancies reflect our inclusion of land and inventories. Appendix tableB1 lists the component assets and 1998 investment and stock values; tableB2 presents the value of the total capital stock and that of each IT asset


25. BEA (1998b).

1.6

1.4

1.2

1.0

0.8

0.6

0.4

0.2

1965 1970 1975 1980 1985 1990 1995

Computer investment

Software investment

Communications investment


Computer and software consumption

Percent

Figure 2. Shares of Information Technology in Total Output, 1960–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. Shares are of current dollar output.


category from 1959 to 1998, as well as asset price indexes for total capi-tal and IT assets.

The stocks of IT business assets (computers, software, and communi-cations equipment), as well as consumers’ purchases of computers andsoftware, have grown dramatically in recent years but remain relativelysmall. In 1998, combined IT assets accounted for only 3.4 percent oftangible capital and 4.6 percent of reproducible private assets.

We now move to estimates of capital services flows, where the capitalstocks of individual assets are aggregated using rental prices as weights.Appendix table B3 presents the current dollar service flows and corre-sponding price indexes for 1959–98, and the second panel of table 1 sum-marizes the growth rates for the prices and quantities of IT inputs for1990–95 and 1995–98.

There is a clear acceleration in the growth of aggregate capital ser-vices, from 2.8 percent per year for 1990–95 to 4.8 percent for 1995–98.This is largely due to rapid growth in services from IT equipment and soft-ware, and it reverses the trend toward slower capital growth through 1995.IT assets account for only 11.2 percent of the total capital service flow, buta much larger share than the corresponding capital stock shares. In 1998capital services were only 12.4 percent of the capital stock for tangibleassets as a whole, but services were 40.0 percent of stocks for IT. Thisreflects the rapid price declines and high depreciation rates that enter intothe rental prices for IT.

Figure 3 highlights the rapid increase in the importance of IT assets,reflecting the accelerating pace of relative price declines. In the 1990s theservice price for computer hardware fell 14.2 percent per year, comparedwith an annual increase of 2.2 percent for non-IT capital. As a direct con-sequence of this relative price change, computer services grew 24.1 per-cent per year, compared with only 3.6 percent for the services of non-ITcapital, in the 1990s. The current dollar share of services from computerhardware increased steadily, to reach nearly 3.5 percent of all capitalservices in 1998 (figure 3).26

The rapid accumulation of software, however, appears to have differ-ent origins. The price of software investment has declined much more


26. Tevlin and Whelan (2000) provide empirical support for this explanation, reportingthat computer investment is particularly sensitive to the cost of capital, so that a rapid dropin service prices can be expected to lead to a large investment response.


slowly, by 1.7 percent per year versus 19.5 percent for computer hard-ware for 1990–98. These differences in investment prices have led to amuch slower decline in service prices for software and computers, 1.6 per-cent per year for software versus 14.2 percent for hardware. Nonetheless,firms have been accumulating software quite rapidly: real capital servicesgrew 13.1 percent per year in the 1990s. Although this is less than the24.1 percent annual growth in computers, software growth has been muchmore rapid than growth in other forms of tangible capital. Complemen-tarity between software and computers is one possible explanation: firmsmay respond to the decline in relative computer prices by accumulatingcomputers and investing in complementary inputs such as software to putthe computers into operation.27

A competing explanation is that the official price indexes used to deflatesoftware investment omit a large part of true quality improvements. Thiswould lead to a substantial overstatement of price inflation and a corre-sponding understatement of real investment, capital services, and eco-


27. An econometric model of the responsiveness of different types of capital services toown- and cross-price effects could be used to test for complementarity, but this is beyond thescope of this paper.

3.5

3.0

2.5

2.0

1.5

1.0

0.5

1965 1970 1975 1980 1985 1990 1995

Percent

Figure 3. Shares of Information Technology in Total Capital and Consumer Durable Services, 1960–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. Shares are of current dollar output.

Communications capital

Computer capital Software capital



nomic growth. According to two recent reports from the BEA,28 onlyprices for prepackaged software are calculated from constant-quality pricedeflators based on hedonic methods. Prices for business own-account soft-ware are based on input cost indexes, which implicitly assume no changein the productivity of computer programmers. Custom software prices area weighted average of prepackaged software and own-account software,with an arbitrary 75 percent weight for business own-account soft-ware prices. Thus the price deflators for nearly two-thirds of recentsoftware investment are estimated under the maintained assumption of nogain in productivity.29 If the quality of own-account and custom softwareis improving at a pace even remotely close to that of packaged software,this implies a large understatement in investment in software.

Although the price decline for communications equipment during the1990s is comparable to that for software, as officially measured in theNIPAs, investment has grown at a rate that is more in line with prices.However, there are also possible measurement biases in the pricing ofcommunications equipment. The technology of switching equipment, forexample, is similar to that of computer hardware; investment in this cate-gory is deflated by a constant-quality price index developed by the BEA.Conventional price deflators are employed for transmission gear, such asfiber-optic cables, which also appear to be declining rapidly in price. Thiscould lead to an underestimate of the rate of growth in communicationsequipment investment, the capital stock, and capital services, as well asan overestimate of the rate of inflation.30 We return to this issue below.

Measuring Labor Services

Here we describe our estimates of labor input for the U.S. economyfrom 1959 to 1998. We begin with individual data from the Census of Pop-ulation for 1970, 1980, and 1990, as well as the annual Current Popula-tion Surveys. We estimate constant-quality indexes for labor input and its


28. Moulton, Parker, and Seskin (1999); Parker and Grimm (2000).29. According to Parker and Grimm (2000), total software investment of $123.4 billion

includes $35.7 billion in prepackaged software, $42.3 billion in custom software, and $45.4billion in own-account software in 1998. Applying the weighting conventions employed bythe BEA, this implies that $46.3 billion = $35.7 billion + 0.25 × $42.3 billion, or 38 per-cent of total software investment, is deflated with explicit quality adjustments.

30. Grimm (1997) presents hedonic estimates for digital telephone switches and reportsaverage price declines of more than 10 percent per year from 1985 to 1996.


price to account for heterogeneity of the work force across sex, employ-ment class, age, and education levels.31

The distinction between labor input and labor hours is analogous tothe distinction between capital services and the capital stock. Growth inlabor input reflects the increase in labor hours, as well as changes in thecomposition of hours worked as firms substitute among heterogeneoustypes of labor. We define growth in labor quality as the difference betweengrowth in labor input and growth in hours worked. Increases in labor qual-ity reflect the substitution of workers with high marginal products for thosewith low marginal products; the growth in hours employed by Solow andothers does not capture this substitution.32 Appendix table C1 presentsour estimates of labor input, hours worked, and labor quality.

Our estimates show a value of labor expenditure of $4.546 trillion in1998, or roughly 57 percent of the value of output. This share accuratelyincludes private output and our imputations for capital services. If weexclude these imputations, labor’s share rises to 62 percent, in line withconventional estimates. As shown in table 1, growth in the index of laborinput Lt appropriate for our model of production in equation 1 acceleratedto 2.8 percent per year for 1995–98, from 2.0 percent for 1990–95. Thisrise was primarily due to growth in hours worked, which rose from 1.4 per-cent per year for 1990–95 to 2.4 percent for 1995–98, as labor forceparticipation increased and unemployment rates plummeted.33

Growth in labor quality decelerated in the late 1990s, from 0.65 percentper year for 1990–95 to 0.43 percent for 1995–98. This slowdown reflectswell-known underlying demographic trends in the composition of thework force, as well as exhaustion of the pool of available workers as unem-ployment rates steadily declined. Projections of future economic growththat omit labor quality, like those of the CBO (discussed below), implicitlyincorporate changes in labor quality into measured TFP growth. Thisreduces the reliability of projections of future economic growth. Fortu-nately, this is easily remedied by extrapolating demographic changes in the


31. This follows the approach of Jorgenson, Gollop, and Fraumeni (1987), whose esti-mates have been revised and updated by Ho and Jorgenson (1999). Appendix C providesdetails on the source data and methodology.

32. Solow (1957).33. By comparison, BLS (2000) reports growth in business hours of 1.2 percent per

year for 1990–95 and 2.3 percent for 1995–98. The slight discrepancies reflect our meth-ods for estimating hours worked by the self-employed, as well as minor differences in thescope of our output measure.



work force in order to reflect foreseeable changes in composition by char-acteristics of workers such as age, sex, and educational attainment.

Quantifying the Sources of Growth

Table 2 presents results of our growth accounting decomposition basedon an extension of equation 2 for the period 1959–98 and various sub-periods, as well as preliminary estimates through 1999. We decomposeeconomic growth by both output and input categories in order to quantifythe contribution of IT to investment and consumption outputs, as well ascapital and consumer durable inputs.34 We extend our previous treatmentof the outputs and inputs of computers by identifying software and com-munications equipment as distinct IT assets.

To quantify the sources of IT-related growth more explicitly, we employan extended production possibilities frontier:

(4)

where outputs include computer and software consumption Cc, computerinvestment Ic, software investment Is, telecommunications investment Im ,the services of consumers’ computers and software Dc, and other outputsYn . Inputs include the capital services of computers Kc , software Ks,telecommunications equipment Km , other capital assets Kn , services ofconsumers’ computers and software Dc, other durables Dn, and labor inputL.35 As in equation 1, TFP is denoted by A and represents the ability to pro-duce more output from the same inputs. Time subscripts have beendropped for convenience.

The corresponding extended growth accounting equation is

(5)

,

w Y w C w Iw I w I w D v K v K

v K v Kv D v Dv L A

Yn n Cc c Ic c

Is s m Dc c Kn n Kc c

Ks s Km m

Dn n Dc c

L

∆ ∆ ∆∆ ∆ ∆ ∆ ∆

∆ ∆∆ ∆

∆ ∆

ln ln lnln ln ln ln ln

ln lnln ln

ln ln

+ ++ + + = +

+ ++ ++ +

Im

, , , , , , , , , , , ,Y Y C I I I D A X K K K K D D Ln c c s m c n c s m n c( ) = ⋅ ( )

34. As in Jorgenson and Stiroh (1999).35. Note that we have broken broadly defined capital into tangible capital services K and

consumer durable services D.


Tab

le 2

. G

row

th in

Pri

vate

Dom

esti

c O

utpu

t an

d th

e So

urce

s of

Gro

wth

, 195

9–99

a

Per

cent

Pre

lim

inar

yVa

riab

le19

59–9

819

59–7

319

73–9

019

90–9

519

95–9

819

95–9

9

Gro

wth

in p

riva

te d

omes

tic

outp

ut (

Y)

3.63

04.

325

3.12

62.

740

4.72

94.

763

Con

trib

utio

n of

sel

ecte

d ou

tput

com

pone

nts

Oth

er (

Yn)

3.27

54.

184

2.78

22.

178

3.65

93.

657

Com

pute

r an

d so

ftw

are

cons

umpt

ion

(Cc)

0.03

50.

000

0.02

30.

092

0.16

70.

175

Com

pute

r in

vest

men

t (I c

)0.

150

0.06

70.

162

0.20

00.

385

0.38

8S

oftw

are

inve

stm

ent (

I s)

0.07

40.

025

0.07

50.

128

0.20

80.

212

Com

mun

icat

ions

inve

stm

ent (

I m)

0.06

00.

048

0.06

10.

053

0.12

20.

128

Com

pute

r an

d so

ftw

are

cons

umer

dur

able

ser

vice

s (D

c)0.

036

0.00

00.

023

0.08

90.

187

0.20

4

Con

trib

utio

n of

cap

ital

ser

vice

s (K

)1.

260

1.43

61.

157

0.90

81.

611

1.72

7O

ther

(K

n)

0.93

61.

261

0.80

70.

509

0.85

70.

923

Com

pute

rs (

Kc)

0.17

70.

086

0.19

90.

187

0.45

80.

490

Sof

twar

e (K

s)0.

075

0.02

60.

071

0.15

40.

193

0.20

5C

omm

unic

atio

ns (

Km)

0.07

30.

062

0.08

00.

058

0.10

40.

109

Con

trib

utio

n of

con

sum

er d

urab

le s

ervi

ces

(D)

0.51

00.

632

0.46

50.

292

0.55

80.

608

Oth

er (

Dn)

0.47

40.

632

0.44

20.

202

0.37

00.

403

Com

pute

rs a

nd s

oftw

are

(Dc)

0.03

60.

000

0.02

30.

089

0.18

70.

204

Con

trib

utio

n of

labo

r (L

)1.

233

1.24

91.

174

1.18

21.

572

1.43

8A

ggre

gate

tota

l fac

tor

prod

uctiv

ity

(TF

P)

0.62

81.

009

0.33

00.

358

0.98

70.

991

Gro

wth

of

capi

tal a

nd c

onsu

mer

dur

able

ser

vice

s4.

212

4.98

53.

847

2.85

14.

935

5.28

6G

row

th o

f la

bor

inpu

t2.

130

2.14

12.

035

2.01

42.

810

2.57

5

Con

trib

utio

n of

cap

ital

and

con

sum

er d

urab

le q

uali

ty0.

449

0.40

20.

405

0.43

40.

945

1.04

1C

ontr

ibut

ion

of c

apit

al a

nd c

onsu

mer

dur

able

sto

ck1.

320

1.66

41.

217

0.76

51.

225

1.29

3C

ontr

ibut

ion

of la

bor

qual

ity

0.31

50.

447

0.20

00.

370

0.25

30.

248

Con

trib

utio

n of

labo

r ho

urs

0.91

80.

802

0.97

40.

812

1.31

91.

190

Ave

rage

labo

r pr

oduc

tivit

y (A

LP

) 2.

042

2.94

81.

437

1.36

62.

371

2.58

0

Sou

rce:

Aut

hors

’cal

cula

tion

s ba

sed

on B

EA

, BL

S, C

ensu

s B

urea

u, a

nd o

ther

dat

a so

urce

s.

a. A

con

trib

utio

n of

an

outp

ut o

r an

inpu

t is

defi

ned

as th

e sh

are-

wei

ghte

d, r

eal g

row

th r

ate.

See

app

endi

xes

A th

roug

h C

for

det

ails

on

esti

mat

ion

and

data

sou

rces

.


where w� and v� denote average shares in nominal income for the sub-scripted variable, w� Yn + w�Cc + w� Ic + w� Is + w�Im + w�Dc = v�Kn + v�Kc + v�Ks + v�Km+ v�Dn + v�Dc + v�L = 1, and we refer to a share-weighted growth rate as thecontribution of an input or output.

OUTPUT GROWTH. We first consider the sources of output growth for theentire 1959–98 period. Broadly defined capital services make the largestgrowth contribution, 1.8 percentage points (1.3 percentage points frombusiness capital and 0.5 percentage point from consumer durable assets);labor services contribute 1.2 percentage points; and TFP growth is respon-sible for only 0.6 percentage point. Input growth is the source of nearly80 percent of U.S. growth over the past forty years, and TFP has accountedfor approximately one-fifth. Figure 4 highlights this result by showingthe relatively small growth contribution of the TFP residual in each sub-period.

More than three-quarters of the contribution of broadly defined capitalreflects the accumulation of capital stock, whereas increased labor hoursaccount for slightly less than three-quarters of labor’s contribution. Thequality of both capital and labor has made important contributions, 0.45 percentage point and 0.32 percentage point per year, respectively.Accounting for substitution among heterogeneous capital and labor inputsis therefore an important part of quantifying the sources of economicgrowth.

A look at the U.S. economy before and after 1973 reveals some famil-iar features of the historical record. After a period of strong output and TFPgrowth in the 1960s and early 1970s, the U.S. economy slowed markedlythrough 1990, with annual output growth falling from 4.3 percent for1959–73 to 3.1 percent for 1973–90, and annual TFP growth falling almosttwo-thirds of a percentage point, from 1.0 percent to 0.3 percent. Growthin capital inputs also slowed, falling from 5.0 percent per year for 1959–73to 3.8 percent for 1973–90, and this contributed to sluggish ALP growth,which fell from 2.9 percent per year for 1959–73 to 1.4 percent for1973–90.

We now focus on the 1990s and highlight some recent changes.36 Rel-ative to the early 1990s, annual output growth increased by nearly


36. Table 2 also presents preliminary results for the more recent period 1995–99, wherethe 1999 numbers are based on the estimation procedure described in appendix E rather thanthe detailed model described above. The results for 1995–98 and 1995–99 are quite simi-lar; we focus our discussion on the period 1995–98.


2 percentage points for 1995–98. The contribution of capital and consumerdurables jumped by 1.0 percentage point, that of labor rose by 0.4 per-centage point, and TFP growth accelerated by 0.6 percentage point. ALPgrowth rose 1.0 percentage point. The rising contributions of labor andcapital encompass several well-known trends in the late 1990s. Growth inhours worked accelerated as labor markets tightened, unemployment fellto a thirty-year low, and labor force participation rates increased.37 Thecontribution of capital reflects the investment boom of the late 1990s, asbusinesses poured resources into plant and equipment, especially com-puters, software, and communications equipment.

The acceleration in TFP growth is perhaps the most remarkable fea-ture of the data. After averaging only 0.34 percent per year from 1973 to1995, TFP accelerated to 0.99 percent per year for 1995–98, suggestingmassive improvements in technology and increases in the efficiency ofproduction. Although the resurgence in TFP growth in the 1990s has yet to


37. Katz and Krueger (1999) explain the strong performance of the U.S. labor marketin terms of demographic shifts toward a more mature labor force, a rise in the prison popu-lation, improved efficiency in labor markets, and the “weak backbone” hypothesis of workerrestraint.

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

1959–73 1973–90 1990–95 1995–98

Percent per year

Total factor productivityb

Labor inputCapital and consumer durable services

Figure 4. Sources of Growth, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. An input’s contribution is its average, share-weighted, annual growth rate.b. Defined in equation 2.


surpass that of the 1960s and early 1970s, more-rapid TFP growth is crit-ical for sustained growth at higher rates.

Figures 5 and 6 highlight the rising contributions of IT outputs to U.S.economic growth. Figure 5 shows the breakdown between IT and non-IToutputs for the same subperiods as in figure 4, whereas figure 6 decom-poses the contribution of IT outputs into the five components identifiedabove, also for these four subperiods. Although the role of IT has steadilyincreased, figure 5 shows that the recent investment and consumptionsurge nearly doubled the output contribution of IT for 1995–98 relative tothat of 1990–95. Figure 6 shows that computer investment was the largestsingle IT contributor in the late 1990s, and that consumption of comput-ers and software is becoming increasingly important as a source of outputgrowth.

Figures 7 and 8 present a similar decomposition of the role of IT assetsas production inputs, where their contribution is rising even more dramat-ically. Figure 7 shows that the contribution of IT to growth in capital andconsumer durables increased rapidly in the late 1990s, and these assetsnow account for more than two-fifths of the total growth contribution frombroadly defined capital. Figure 8 shows that computer hardware is also thesingle largest IT contributor on the input side, reflecting the growing shareand rapid growth rates of the late 1990s.

The contribution of computers, software, and communications equip-ment presents a different picture from that depicted in our previousresearch, for both data and methodological reasons.38 First, the BEAbenchmark revision has classified software as an investment good.Although software is growing more slowly than computers, the substantialnominal share of software services has raised the overall contribution ofIT. Second, we have added communications equipment, also a slower-growing component of capital services, with similar effects. Third, we nowincorporate asset-specific revaluation terms in all rental price estimates.Since the acquisition prices of computers are falling steadily, asset-specificrevaluation terms have raised the estimated service price and increased theshare of computer services. Finally, we have modified our timing conven-tion and now assume that capital services from individual assets are pro-portional to the average of the current and the lagged stock. For assets withrelatively short service lives like most IT, this is a more reasonable


38. Jorgenson and Stiroh (1999).


4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

0.5

1959–73 1973–90 1990–95 1995–98

Percent per year

IT outputNon-IT output

Figure 5. Contribution of Information Technology to Output, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. An output’s contribution is its average, share-weighted, annual growth rate.

1.0

0.8

0.6

0.4

0.2

1959–73 1973–90 1990–95 1995–98

Percent per year

Computer and software consumer durable servicesCommunications investmentSoftware investmentComputer investmentComputer and software consumption

Figure 6. Contribution of Information Technology to Output, by Type of Asset, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. An output’s contribution is its average, share-weighted, annual growth rate.



assumption than in our earlier work, which assumed that it took a fullyear for new investment to become productive.39

This large increase in the growth contribution of computers and soft-ware is consistent with recent estimates by Stephen Oliner and DanielSichel,40 although their estimate of the contribution is somewhat larger.They report that computer hardware and software contributed 0.95 per-centage point to growth for 1996–99, whereas communications con-tributed another 0.15 percentage point. The discrepancy between theirestimates and ours primarily reflects our broader output concept, whichlowers the input share of these high-technology assets, as well as minordifferences in tax parameters and stock estimates. Karl Whelan alsoreports a larger growth contribution (0.82 percentage point) from computerhardware for 1996–98.41 The discrepancy again reflects our broader output


39. We are indebted to Daniel Sichel for very helpful discussions of this timingconvention.

40. Oliner and Sichel (2000).41. Whelan (2000).

2.0

1.5

1.0

0.5

1959–73 1973–90 1990–95 1995–98

Percent per year

IT capital and consumer durable servicesNon-IT capital and consumer durable services

Figure 7. Contribution of Information Technology to Capital and Consumer Durable Input, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. An input’s contribution is its average, share-weighted, annual growth rate.


concept. In addition, Whelan introduces a new methodology to account forretirement and support costs; this methodology generates a considerablylarger capital stock and raises the input share and the growth contributionfrom computer capital.

Despite differences in methodology and data sources among studies, aconsensus is building that computers are having a substantial impact oneconomic growth.42 What is driving the increase in the contributions ofcomputers, software, and communications equipment? As we have arguedelsewhere,43 price changes lead to substitution toward capital services withlower relative prices. Firms and consumers are responding to relative pricechanges.

Table 1 showed that the acquisition price of computer investment fellnearly 28 percent per year, the price of software fell 2.2 percent per year,and the price of communications equipment fell 1.7 percent per year during


42. Oliner and Sichel (2000) provide a detailed comparison of the results across severalstudies of computers and economic growth.

43. Jorgenson and Stiroh (1999).

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

1959–73 1973–90 1990–95 1995–98

Percent per year

Computer and software consumer durablesCommunications capitalSoftware capitalComputer capital

Figure 8. Contribution of Information Technology to Capital and Consumer Durable Input, by Type of Asset, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. An input’s contribution is its average, share-weighted, annual growth rate.


the period 1995–98, while other output prices rose 2.0 percent per year. Inresponse to these price changes, firms accumulated computers, software,and communications equipment more rapidly than other forms of capital.Investment other than in IT actually declined as a proportion of privatedomestic product. The story of household substitution toward computersand software is similar. These substitutions suggest that the gains of thecomputer revolution accrue to firms and households that are adept atrestructuring activities to respond to these relative price changes.

AVERAGE LABOR PRODUCTIVITY GROWTH. To provide a different per-spective on the sources of economic growth, we focus next on ALPgrowth. By simple arithmetic, output growth equals the sum of growth inhours and growth in labor productivity.44 Table 3 shows the output break-down between growth in hours and ALP for the same pre-1999 periods asin table 2. For the entire period 1959–98, ALP growth was the predomi-nant determinant of output growth, increasing just over 2 percent per year,while hours increased about 1.6 percent per year. We then examine thechanging importance of the factors determining ALP growth. As shownin equation 3, ALP growth depends on a capital deepening effect, a laborquality effect, and a TFP effect.

Figure 9 plots the importance of each factor, revealing the well-knownproductivity slowdown of the 1970s and 1980s and highlighting the accel-eration of labor productivity growth in the late 1990s. The slowdownthrough 1990 reflects declines in all three components: less capital deep-ening, declining labor quality growth, and decelerating growth in TFP. Thegrowth of ALP slipped further during the early 1990s, with the seriousslump in capital deepening only partly offset by a revival in the growth oflabor quality and an uptick in TFP growth. Slow growth in hours combinedwith slow ALP growth during 1990–95 to produce a further slide in thegrowth of output. This stands out from previous cyclical recoveries dur-ing the postwar period, when output growth accelerated during the recov-ery, powered by more rapid growth in hours and in ALP.

For the most recent period, 1995–98, strong output growth reflectsgrowth in labor hours and in ALP almost equally. Comparing 1990–95with 1995–98, output growth accelerated by nearly 2 percentage points


44. See Krugman (1997) and Blinder (1997) for discussions of the usefulness of thisrelationship.


as a result of a 1.0-percentage-point increase in hours worked and a1.0-percentage-point increase in ALP growth.45 Figure 9 shows that theacceleration in ALP growth was due to rapid capital deepening from theinvestment boom, as well as faster TFP growth. Capital deepening con-tributed 0.49 percentage point to the acceleration in ALP growth, while


45. BLS data show similar trends for the business sector, with hours growth increasingfrom 1.2 percent per year during 1990–95 to 2.3 percent per year during 1995–98, while

Table 3. Sources of Growth in Average Labor Productivity, 1959–98Percent

Variable 1959–98 1959–73 1973–90 1990–95 1995–98

Growth of private domestic output (Y) 3.630 4.325 3.126 2.740 4.729Growth in hours (H) 1.588 1.377 1.689 1.374 2.358Growth in ALP (Y/H) 2.042 2.948 1.437 1.366 2.371

Contribution of capital deepening 1.100 1.492 0.908 0.637 1.131to ALPa

Contribution of labor quality 0.315 0.447 0.200 0.370 0.253to ALP

Contribution of TFP to ALP 0.628 1.009 0.330 0.358 0.987

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources. a. ALP contributions are defined in equation 3.

3.0

2.5

2.0

1.5

1.0

0.5

1959–73 1973–90 1990–95 1995–98

Percent per year

Total factor productivityb

Labor qualityCapital deepening

Figure 9. Sources of Labor Productivity Growth, 1959–98a

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources.a. Contributions are defined in equation 3.b. Defined in equation 2.


acceleration in TFP growth added 0.63 percentage point. Growth in laborquality slowed somewhat as growth in hours accelerated. This reflects thefalling unemployment rate and tightening of labor markets, as more work-ers with relatively low marginal products were drawn into the work force.46

Our decomposition also throws some light on the hypothesis advancedby Gordon, who argues that the vast majority of recent ALP gains are dueto the production of IT, particularly computers, rather than the use of IT.47

As we have already pointed out, more efficient IT production generatesaggregate TFP growth as more computing power is produced from thesame inputs, whereas IT use affects ALP growth through capital deepen-ing. In recent years, acceleration of TFP growth has been a slightly moreimportant factor in the acceleration of ALP growth than capital deepening.Efficiency gains in computer production are an important part of aggregateTFP growth, as Gordon’s results on ALP suggest. We discuss this issue ingreater detail below.

TOTAL FACTOR PRODUCTIVITY GROWTH. Finally, we consider theremarkable performance of U.S. TFP growth in recent years. After main-taining an average annual rate of 0.33 percent for the period 1973–90, TFPgrowth rose to 0.36 percent per year for 1990–95 and then vaulted to 0.99percent per year for 1995–98. This jump is a major source of growth inoutput and ALP for the U.S. economy (figures 4 and 9). Although TFPgrowth for the 1990s has yet to reattain its peaks of certain periods in thegolden age of the 1960s and early 1970s, the recent acceleration suggeststhat the U.S. economy may be recuperating from the anemic productivitygrowth of the past two decades. Of course, caution is warranted until morehistorical experience is available.

Since Domar’s 1961 article, economists have utilized a multi-industrymodel of the economy to trace aggregate TFP growth to its sources at thelevel of individual industries. Jorgenson and colleagues have employedthis model to identify industry-level sources of growth.48 More recently,William Gullickson and Michael Harper, and Jorgenson and Stiroh, have


ALP increased from 1.6 percent per year to 2.6 percent per year. Data are available atwww.bls.gov.

46. Oliner and Sichel (2000) also show a decline in the growth contribution of laborquality in the late 1990s, from 0.44 percentage point during 1991–95 to 0.31 percentagepoint during 1996–99.

47. Gordon (1999b).48. Jorgenson, Gollop, and Fraumeni (1987); Jorgenson (1990).


used the model for similar purposes.49 We postpone more detailed consid-eration of the sources of TFP growth until we have examined the impact ofalternative price deflators on our growth decomposition.

Alternative Growth Accounting Estimates

Tables 1 through 3 and figures 1 through 9 reported our primary resultsusing the official data published in the NIPAs. As already noted, however,there is reason to believe that the rates of inflation in official price indexesfor certain high-technology assets, notably software and telecommunica-tions equipment, may be overstated. Two recent papers from the BEA,for example, report that only the prepackaged portion of software invest-ment is deflated with a constant-quality deflator.50 Own-account softwareis deflated with an input cost index, and custom software is deflated with aweighted average of the prepackaged and own-account deflator. Simi-larly, the BEA reports that, in the communications equipment category,only telephone switching equipment is deflated with a constant-quality,hedonic deflator.51

Here we incorporate alternative price series for software and commu-nications equipment and examine the impact on estimates of U.S. eco-nomic growth and its sources. Table 4 presents growth accounting resultsunder three different scenarios. The “base case” repeats the estimates fromtable 2, which are based on official NIPA price data. Two additional casesincorporate price series for software and communications equipment thatshow faster price declines and correspondingly more rapid real investmentgrowth.52

The “moderate price decline” case assumes that prepackaged softwareprices are appropriate for all types of private software investment, includ-ing custom and business own-account software. Since the index forprepackaged software is based on explicit quality adjustments, it fallsmuch faster than the prices of custom and own-account software: 10.1 per-cent per year versus 0.4 percent and 4.1 percent per year, respectively, for


49. Gullickson and Harper (1999); Jorgenson and Stiroh (2000).50. Moulton, Parker, and Seskin (1999); Parker and Grimm (2000).51. Bruce Grimm, Bureau of Economic Analysis, personal communication.52.The notion that official price deflators for investment goods omit substantial quality

improvements is hardly novel. The magisterial work of Gordon (1990) successfully quanti-fied the overstatements of rates of inflation for the prices of a wide array of investmentgoods, covering all producer durable equipment in the NIPAs.


Tab

le 4

. Im

pact

of A

lter

nati

ve D

eflat

ions

of

Soft

war

e an

d C

omm

unic

atio

ns E

quip

men

t on

the

Sou

rces

of

Gro

wth

, 195

9–98

a

Per

cent

Bas

e ca

seM

oder

ate

pric

e de

clin

eR

apid

pri

ce d

ecli

ne

Vari

able

1959

–73

1973

–90

1990

–95

1995

–98

1959

–73

1973

–90

1990

–95

1995

–98

1959

–73

1973

–90

1990

–95

1995

–98

Gro

wth

in p

riva

te d

omes

tic

outp

ut (

Y)

4.33

3.13

2.74

4.73

4.35

3.30

2.90

4.89

4.36

3.38

3.03

5.07

Con

trib

utio

n of

sel

ecte

d ou

tput

com

pone

nts

Oth

er (

Yn)

4.18

2.78

2.18

3.66

4.12

2.76

2.17

3.66

4.08

2.75

2.16

3.66

Com

pute

r an

d so

ftw

are

cons

umpt

ion

(Cc)

0.00

0.02

0.09

0.17

0.00

0.02

0.09

0.17

0.00

0.02

0.09

0.17

Com

pute

r in

vest

men

t (I c

)0.

070.

160.

200.

390.

070.

160.

200.

390.

070.

160.

200.

39S

oftw

are

inve

stm

ent (

I s)

0.03

0.08

0.13

0.21

0.04

0.14

0.22

0.29

0.05

0.17

0.29

0.40

Com

mun

icat

ions

inve

stm

ent (

I m)

0.05

0.06

0.05

0.12

0.12

0.19

0.13

0.21

0.16

0.25

0.19

0.27

Com

pute

r an

d so

ftw

are

cons

umer

dur

able

0.

000.

020.

090.

190.

000.

020.

090.

190.

000.

020.

090.

19se

rvic

es (

Dc)

Con

trib

utio

n of

cap

ital

ser

vice

s (K

)1.

441.

160.

911.

611.

541.

391.

151.

831.

611.

511.

322.

09O

ther

(K

n)

1.26

0.81

0.51

0.86

1.25

0.80

0.51

0.86

1.25

0.79

0.51

0.85

Com

pute

rs (

Kc)

0.09

0.20

0.19

0.46

0.09

0.20

0.19

0.46

0.09

0.20

0.19

0.46

Sof

twar

e (K

s)0.

030.

070.

150.

190.

050.

150.

280.

290.

060.

180.

360.

45C

omm

unic

atio

ns (

Km)

0.06

0.08

0.06

0.10

0.16

0.25

0.18

0.23

0.22

0.34

0.27

0.33

Con

trib

utio

n of

con

sum

er d

urab

le s

ervi

ces

(D)

0.63

0.47

0.29

0.56

0.63

0.46

0.29

0.56

0.63

0.46

0.29

0.56

Oth

er (

Dn)

0.63

0.44

0.20

0.37

0.63

0.44

0.20

0.37

0.63

0.44

0.20

0.37

Com

pute

rs a

nd s

oftw

are

(Dc)

0.00

0.02

0.09

0.19

0.00

0.02

0.09

0.19

0.00

0.02

0.09

0.19

Con

trib

utio

n of

labo

r (L

)1.

251.

171.

181.

571.

251.

171.

181.

571.

251.

181.

181.

57A

ggre

gate

tota

l fac

tor

prod

uctiv

ity

1.01

0.33

0.36

0.99

0.94

0.27

0.27

0.93

0.88

0.22

0.23

0.85

Gro

wth

of

capi

tal a

nd c

onsu

mer

dur

able

ser

vice

s4.

993.

852.

854.

945.

244.

403.

435.

445.

414.

703.

846.

02G

row

th o

f la

bor

inpu

t2.

142.

042.

012.

812.

142.

042.

012.

812.

142.

042.

012.

81

Con

trib

utio

n of

cap

ital

and

con

sum

er d

urab

le q

uali

ty0.

400.

410.

430.

950.

480.

590.

631.

110.

540.

700.

781.

34C

ontr

ibut

ion

of c

apit

al a

nd c

onsu

mer

dur

able

sto

ck1.

661.

220.

771.

231.

681.

260.

821.

281.

691.

270.

841.

31C

ontr

ibut

ion

of la

bor

qual

ity

0.45

0.20

0.37

0.25

0.45

0.20

0.37

0.25

0.45

0.20

0.37

0.25

Con

trib

utio

n of

labo

r ho

urs

0.80

0.97

0.81

1.32

0.80

0.97

0.81

1.32

0.80

0.98

0.81

1.32

Ave

rage

labo

r pr

oduc

tivit

y 2.

951.

441.

372.

372.

981.

611.

522.

532.

991.

691.

652.

72

Sou

rce:

Aut

hors

’cal

cula

tion

s ba

sed

on B

EA

, BL

S, C

ensu

s B

urea

u, a

nd o

ther

dat

a so

urce

s.

a. T

he b

ase

case

use

s of

ficia

l pri

ce d

ata

from

the

NIP

As.

The

mod

erat

e pr

ice

decl

ine

case

use

s th

e de

flat

or fo

r pre

pack

aged

sof

twar

e fo

r all

sof

twar

e an

d an

nual

pri

ce c

hang

es o

f –10

.7 p

erce

nt fo

r com

mun

icat

ions

equ

ip-

men

t. T

he r

apid

pri

ce d

ecli

ne c

ase

uses

ann

ual p

rice

cha

nges

of

–16

perc

ent f

or s

oftw

are

and

–17.

9 pe

rcen

t for

com

mun

icat

ions

equ

ipm

ent.

See

the

text

for

det

ails

. A c

ontr

ibut

ion

is d

efine

d as

the

shar

e-w

eigh

ted,

rea

lgr

owth

rat

e.


the full 1959–98 period.53 For communications equipment the data aremore limited, and we assume that prices fell 10.7 percent per year through-out the entire period. This estimate is the average annual “smoothed”decline for digital switching equipment for 1985–96 reported by BruceGrimm.54 Although this series may not be appropriate for all types of com-munications equipment, it exploits the best available information.

The “rapid price decline” case assumes that software prices fell 16 per-cent per year for 1959–98, the rate of quality-adjusted price declinereported by Erik Brynjolfsson and Chris Kemerer for microcomputerspreadsheets for 1987–92.55 This is a slightly faster decline than the 15 per-cent annual rate for 1986–91 estimated by Neil Gandal, and considerablyfaster than the 3 percent annual decline for word processors, spreadsheets,and databases for 1987–93 reported by Oliner and Sichel.56 For commu-nications equipment we used estimates from the most recent period fromGrimm,57 who reports a decline of 17.9 percent per year for 1992–96.

Although this exercise necessarily involves some arbitrary choices, theestimates incorporate the limited data now available and provide a valu-able perspective on the crucial importance of accounting for qualitychange in the prices of investment goods. Comparisons among the threecases are also useful in suggesting the range of uncertainty currently con-fronting analysts of U.S. economic growth.

Before discussing the empirical results, it is worthwhile to emphasizethat a more rapid price decline for IT has two direct effects on the sourcesof growth and one indirect effect. The alternative investment deflators raisemeasured real output growth by reallocating nominal growth away fromprices and toward quantities. This also increases the growth rate of thecapital stock, since there are larger investment quantities in each year.More-rapid price declines also give greater weight to capital services from IT.

The counterbalancing effects of increased output growth and increasedinput growth lead to an indirect effect on measured TFP growth. Depend-ing on the relative shares of high-technology assets in investment and


53. Parker and Grimm (2000).54. Grimm (1997).55. Brynjolfsson and Kemerer (1996).56. Gandal (1994); Oliner and Sichel (1994).57. Grimm (1997).


capital services, the TFP residual will increase if the output effect domi-nates and decrease if the effect on capital services dominates.58 Green-wood, Hercowitz, and Krusell, following Solow, omit the output effect andattribute the input effect to “investment-specific” (embodied) technicalchange.59 This must be carefully distinguished from the effects of industry-level productivity growth on TFP growth, discussed below.

Table 4 reports growth accounting results from these three scenarios: thebase case, the moderate price decline case, and the rapid price decline case.The results are not surprising: the more rapid the price decline for softwareand communications, the faster the rate of growth of output and capital ser-vices. Relative to the base case, output growth increases by 0.16 percentagepoint per year for 1995–98 in the moderate price decline case and by0.34 percentage point per year in the rapid price decline case. Capital inputgrowth shows slightly larger increases across the three cases. Clearly,constant-quality price indexes for IT are essential for further progress inunderstanding the growth impact of high-technology investment.

The acceleration in output and input growth reflects the increased con-tributions from IT and determines the effect on the TFP residual. In par-ticular, the output contribution from software for 1995–98 increases from0.21 percentage point in the base case, to 0.29 percentage point in themoderate price decline case, to 0.40 percentage point in the rapid pricedecline case. Similarly, the capital services contribution for softwareincreases from 0.19 to 0.29 to 0.45 percentage point. The contribution ofcommunications equipment shows similar changes. Residual TFP growthfalls slightly during the 1990s, as the input effect outweighs the outputeffect because of the large capital services shares of IT.

This exercise illustrates the sensitivity of the sources of growth to alter-native price indexes for IT. We do not propose to argue that either of the twoalternative cases is more nearly correct than the base case with the officialprices from the NIPAs. Given the paucity of quality-adjusted price data onhigh-technology equipment, we simply do not know. Rather, we have triedto highlight the importance of correctly measuring prices and quantities tounderstand the dynamic forces driving U.S. economic growth. As high-technology assets continue to proliferate through the economy and as other


58. This point was originally made by Jorgenson (1966); Hulten (2000) provides a recentreview.

59. Solow (1957); Greenwood, Hercowitz, and Krusell (1997).


investment goods become increasingly dependent on electronic compo-nents, these measurement issues will become increasingly important.Although the task that lies ahead is onerous, the creation of quality-adjustedprice indexes for all high-technology assets deserves top priority.

Decomposition of TFP Growth

We next consider the role of high-technology industries as a source ofTFP growth. As discussed above, production of high-technology invest-ment goods has made important contributions to aggregate growth. TheCEA, for example, allocates 0.39 percentage point of aggregate TFPgrowth to computer production, and Oliner and Sichel allocate 0.49 per-centage point to the production of computers and computer-related semi-conductor production for the period 1995–99.60

We employ a methodology based on the price “dual” approach to mea-surement of productivity at the industry level. Anticipating our completeindustry analysis below, it is worthwhile to spell out the decompositionof TFP growth by industry. Using the Domar approach to aggregation, weweight industry-level productivity growth by the ratio of the gross outputof each industry to aggregate value added to estimate the industry’s con-tribution to aggregate TFP growth. In the dual approach, the rate of pro-ductivity growth is measured as the decline in the price of output, plus aweighted average of the growth rates of input prices.

In the case of computer production, this expression is dominated by twoterms: the price of computers and the price of semiconductors, which are aprimary intermediate input into the computer-producing industry. If semi-conductor industry output is used only as an intermediate good to pro-duce computers, then its contribution to computer industry productivitygrowth, weighted by computer industry output, precisely cancels its inde-pendent contribution to aggregate TFP growth.61 This independent contri-bution from the semiconductor industry, based on the complete Domarweighting scheme, is the value of semiconductor output divided by aggre-gate value added, multiplied by the rate of price decline in semiconductors.


60. CEA (2000); Oliner and Sichel (2000). Gordon (1999a), Stiroh (1998a), and Whelan(1999) have also provided estimates.

61. This calculation shows that the simplified model of Oliner and Sichel (2000) is a spe-cial case of the complete Domar weighting scheme used below.


Table 5 reports details of our TFP decomposition for the three alterna-tive cases described above for 1990–95 and 1995–98, and figure 10 sum-marizes the IT versus non-IT comparison. In our base case, using offi-cial NIPA data, we estimate that the production of IT accounts for 0.44 percentage point of TFP growth for 1995–98, compared with 0.25 percentage point for 1990–95. This reflects the accelerating relativeprice changes due to radical shortening of the product cycle forsemiconductors.62


62. Relative price changes in the base case are taken from the investment prices in table 5. Output shares are estimated based on final demand sales for computers, availablefrom the BEA website (www.bea.doc.gov/bea/dnl.htm) and from Parker and Grimm (2000)for software. Investment in communications equipment is from the NIPAs, and we esti-mate other final demand components for communications equipment using ratios relativeto final demand for computers. This is an approximation necessitated by the lack of com-plete data on sales to final demand by detailed commodity.

Table 5. Information Technology Decomposition of Total Factor ProductivityGrowth for Alternative Deflation Cases, 1990–98a

Moderate RapidBase case price decline price decline

Variable 1990–95 1995–98 1990–95 1995–98 1990–95 1995–98

Aggregate TFP growth 0.36 0.99 0.27 0.93 0.23 0.85

TFP contributionInformation technology 0.25 0.44 0.46 0.64 0.64 0.86

Computers 0.16 0.32 0.16 0.32 0.16 0.32Software 0.05 0.08 0.17 0.18 0.28 0.34Communications 0.04 0.04 0.13 0.13 0.21 0.20

Non–information technology 0.11 0.55 –0.19 0.29 –0.41 –0.01

Relative price changeComputers –16.6 –29.6 –16.6 –29.6 –16.6 –29.6Software –3.4 –4.2 –11.3 –9.7 –18.0 –18.0Communications –3.5 –3.8 –12.7 –12.7 –19.9 –19.9

Average nominal shareComputers 0.96 1.09 0.96 1.09 0.96 1.09Software 1.54 1.88 1.54 1.88 1.54 1.88Communications 1.05 1.02 1.05 1.02 1.05 1.02

Source: Authors’ calculations based on BEA, BLS, Census Bureau, and other data sources. a. The base case uses official price data

Documents

Raising the Speed Limit: U.S. Economic Growth in the ......1990–91 recession (Bureau of Labor Statistics, 2000). ... plays as a source of economic growth. 3 For the period 1959–73,